An application of Neural Network for Human Performance Modelling

Human Performance Modelling (HPM) is needed in many applications e.g. predicting driver performance for semi-autonomous vehicles, eSports gamers, adaptive machines in manufacturing to improve safety, productivity or a response from a machine. However, it is often challenging to establish a reliable HPM. This paper proposes and validates Neural Network (NN) models for human performance using a simulated Inverted Pendulum Driven Capsule Model (IPDCM) platform as a case study. The joystick data is utilised to build a classification and a regression model to predict the IPDCM’s movement direction and position, respectively. Then the number of neurons in the single hidden layer is optimised for both models using 10-time-10-fold cross-validation. Finally, the optimised classification and regression NN models are applied to a blind test dataset which exhibits accuracies of 92.2% and 77.01%, respectively.